Method For Recognizing And Localizing Actions In Compressed Video
- Technology Benefits
- Low computational costsAppearance invariantDoesn't require prior video segmentation, alignment, stabilization or decompression
- Technology Application
- Real-time recognition and localization of actions in compressed video data.
- Detailed Technology Description
- None
- Supplementary Information
- Patent Number: US8027542B2
Application Number: US2007764754A
Inventor: Ahammad, Parvez | Yeo, Chuohao | Ramchandran, Kannan | Sastry, S. Shankar
Priority Date: 18 Jun 2007
Priority Number: US8027542B2
Application Date: 18 Jun 2007
Publication Date: 27 Sep 2011
IPC Current: G06K000948 | G06K000900 | G06K000936 | H03F000126 | H04N000514 | H04N000712
US Class: 382197 | 348699 | 37524016 | 382107 | 382236 | 702189
Assignee Applicant: The Regents of the University of California
Title: High speed video action recognition and localization
Usefulness: High speed video action recognition and localization
Summary: System for detecting and locating an action in a video from the applications. Uses include but are not limited to walking, running, jumping, throwing a punch and action performed by an animate and inanimate actor from video surveillance application, collision-avoidance application for unmanned aerial vehicles, video editing application, video piracy prevention application and viewing application.
Novelty: Action e.g. walking, detecting and locating system for use in e.g. video surveillance application, has employing unit employing similarity value to recognize action or similar action, which occurs in query video and test video
- Industry
- ICT/Telecom
- Sub Category
- Image Processing
- Application No.
- 8027542
- Others
-
Related Technologies
Tech ID/UC Case
17745/2007-021-0
Related Cases
2007-021-0
- *Abstract
-
As their cost have decreased, video camera usage has proliferated in consumer, commercial and government applications. This growth in video usage is leading to an increased need for tools that can perform rudimentary action recognition on compressed video in order to assist users in focusing on targeted actions within voluminous video footage.
To address this opportunity, researchers at UC Berkeley have developed a computational method that can recognize and localize targeted actions in compressed video data. This real-time method uses compressed domain features that can be extracted with low complexity. This novel technique can localize actions in both space and time. Moreover, it is appearance invariant, and it doesn't require prior segmentation, alignment, stabilization or decompression.
The Berkeley research team tested their method on a large action video database consisting of six actions performed by 25 people under 3 different scenarios. The results compared favorably to existing methods at only a fraction of the computational costs of those previous methods.
- *IP Issue Date
- Sep 27, 2011
- *Principal Investigator
-
Name: Parvez Ahammad
Department:
Name: Kannan Ramchandran
Department:
Name: Sosale Sastry
Department:
Name: Chuohao Yeo
Department:
- Country/Region
- USA
